Multivariate biological profiling and principal toxicity regions of compounds: the PCB case study
2002 (English)In: Journal of Chemometrics, Vol. 16, no 8-10, 497-509 p.Article in journal (Refereed) Published
The chemometric QSAR strategy, as applied in environmental sciences and drug design, is based on (1) multivariate characterization of chemical structure, (2) multivariate design in the principal properties of a set of compounds to select a representative training set, and (3) multivariate modelling of the structure-activity relationships. A multivariate QSAR investigation is often commenced by applying a screening design, and the selected compounds are tested biologically in a broad battery of test systems (multivariate biological profiling). In many cases the result is such that for certain biological end-points only some of the tested compounds are active, while for another set of biological end-points other tested chemicals are active. In other words, when looking at the chemical property space, there may be both responding and non-responding toxicity regions, or even regions of very specific toxicity mechanisms. This may lead to loss of resolution and balance in the resulting QSAR models. Therefore it might sometimes be worthwhile to focus the QSAR modelling on parts of the chemical space where high toxicity is expected or known to be the case. In this paper we describe a multi-stage modification of the chemometric QSAR strategy, aimed at identifying focused sets of compounds that provide a good mapping of such principal toxicity regions. This strategy is based on PCA, PLS and multivariate design in several stages. The strategy is illustrated using a data set of polychlorinated biphenyls, a set of compounds for which seven biological end-points were determined. Copyright © 2002 John Wiley & Sons, Ltd.
Place, publisher, year, edition, pages
2002. Vol. 16, no 8-10, 497-509 p.
multivariate biological profiling, principal toxicity regions, multivariate design, PCA, PLS
IdentifiersURN: urn:nbn:se:umu:diva-8807DOI: doi:10.1002/cem.753OAI: oai:DiVA.org:umu-8807DiVA: diva2:148478